US Lifts Export Ban on Anthropic's Fable 5 and Mythos 5 AI Models

A Tale of Two Models: One Restored, One Revealed

Well, that was a ride. Anthropic just pulled off a whiplash reversal for the ages. The AI safety darling announced Wednesday it’s restoring global access to its heavyweight Claude Fable 5 and Mythos 5 models. This comes after a dramatic, weeks-long shutdown, one forced by a U.S. government export control order. But that's not all. In the same breath, the company unveiled Claude Sonnet 5, a brand-new model pitched as an affordable workhorse that it says nips at the heels of its top-tier Opus 4.8.

A dizzying double-feature. One moment, Anthropic is in a public standoff with Washington over national security. The next, it’s shipping a powerful new model for the masses. What gives? These twin announcements show a company scrambling to navigate immense regulatory heat while sprinting to compete in a market that absolutely refuses to stand still.

The Fable and the Fury

The whole saga kicked off just three days after Fable 5 launched on June 9, 2026. Barely out of the gate. Citing national security risks, the U.S. Department of Commerce dropped a bombshell: an unprecedented export control directive, sent straight to Anthropic CEO Dario Amodei. The order? Immediately block all foreign nationals from accessing Fable 5 and the even more potent Mythos 5.

The government’s fear was that the models’ sophisticated cybersecurity skills could be weaponized by foreign adversaries for devastating cyberattacks. This wasn't some abstract worry, either. The concern was supercharged by a recent 'Five Eyes' intelligence alliance warning that advanced AI will “fundamentally transform” cyber capabilities within months, not years. An Amazon research paper, reportedly showing a way to bypass Fable's safeguards, seems to have been the final straw.

So what could Anthropic do? It couldn't reliably check every user's nationality in real-time. The company took the only step it could: disabling the models for everyone. For everyone. This, of course, sparked a firestorm in the AI community. For over two weeks, one of the industry's most advanced AIs just went dark.

Not anymore. The controls are gone, following what Commerce Secretary Howard Lutnick called close work to "analyze and approve Fable 5." White House Chief of Staff Susie Wiles echoed that collaborative tone: “Our shared priority remains: get the best tech deployed as quickly and safely as possible.” To make it happen, Anthropic says it rolled out a new safety classifier that stops the problematic 'jailbreak' technique in over 99% of attempts. But this resolution came with strings. Big ones. Anthropic has now agreed to proactively hunt for security risks, work with the U.S. government on future model releases, and report any malicious use. While Fable 5 is back for the world, the story for the ultra-powerful Mythos 5 is a bit more complicated—it saw a partial restoration to a handful of vetted US organizations on June 26 before the full lift and remains restricted to a select group under its Project Glasswing initiative for defending critical infrastructure.

Enter Sonnet 5: The Workhorse

Just as the dust was settling from the Fable drama, Anthropic pushed its other big news. Say hello to Claude Sonnet 5. The company is billing this one as its 'most agentic' mid-tier model, engineered for complex, multi-step tasks involving browsers and terminals. Here's the pitch: you get performance that’s getting awfully close to the flagship Opus 4.8 model, but for a whole lot less money.

And the pricing? Aggressive. Sonnet 5 is launching with an introductory rate of just $2 per million input tokens and $10 per million output tokens. That sweet deal lasts until August 31, 2026, when prices bump up to $3 and $15. For comparison, the flagship Opus 4.8 costs $5 per million input and a hefty $25 per million output. The math makes this a seriously compelling option for developers building AI agents. Benchmarks show Sonnet 5 absolutely trouncing its predecessor, Sonnet 4.6, while narrowing the gap with Opus 4.8 on most fronts—though Opus still holds a clear lead in the most complex reasoning and coding.

What Does This Mean for AI Regulation?

The Fable 5 shutdown was a first. A big one. It marked the first time U.S. export controls were slapped on AI software itself, not just the silicon it runs on. The incident, and its resolution, sets a massive precedent. We're looking at a new chapter of direct government oversight for any frontier AI model deemed to have dual-use capabilities. As Box CEO Aaron Levie pointed out on X, this outcome could become the template for reviewing future models with advanced cyber skills, demanding "heavy judgment and back and forth between labs and the government for major releases."

For its part, Anthropic is now calling for a shared industry standard to assess and triage potential AI model jailbreaks. The company’s argument is simple: without an agreed-upon way to classify a jailbreak's severity, the entire industry is flying blind. The Fable 5 affair was a live-fire drill in AI governance. Its aftershocks will shape policy for years to come. Anthropic might have navigated this particular crisis, but make no mistake—the days of dropping powerful new models without a government sign-off are over.